MadaneA / MTS-CGAN

Transformer-based Conditional Generative Adversarial Network for Multivariate Time Series Generation (IWTA - PAKDD2023)

MTS-CGAN

Transformer-based Conditional Generative Adversarial Network for Multivariate Time Series Generation

This repository contains the code for the MTS-CGAN model, developed for an article presented at the International Workshop on Temporal Analytics @PAKDD 2023. The model is designed for the conditional generation of realistic multivariate time series data.

Table of Contents

• Quick Start
• Requirements
• Acknowledgments
• Citation

Quick Start

The implementation is divided into several scripts:

• dataLoader.py: Downloads and loads the benchmark dataset used in the paper (UniMiB SHAR).
• MTSCGAN.py: Creates the CGAN model.
• functions.py: Contains functions used to train the model.
• train_MTSCGAN.py: Main script to train the model.
• MTSCGAN_Train.py: Contains the parameter configuration used to train the model.
• LoadSyntheticdata.py: Generates synthetic data using the MTS-CGAN model.
• FID.py: Contains the function used to compute the Frechet Inception Distance (FID).
• DTW.py: Contains the function used to compute the Dynamic Time Warping (DTW) metric.

To train the MTS-CGAN model using the configuration parameters in MTSCGAN_Train.py, use the following command:

$ python3 MTSCGAN_Train.py

The dataset is downloaded automatically.

Training generates several outputs:

• A folder containing a log of training metrics and the model weights
• A folder containing a checkpoint of the model
• A folder containing generated samples

Note: For training, an NVIDIA GPU is strongly recommended for speed. CPU is supported but training is very slow.

Requirements

The main dependencies are:

• torch
• numpy
• pandas
• matplotlib

Acknowledgments

This implementation is based on the open-source code from TransGAN and TTSGAN. We would like to express our gratitude for their contribution to the research community.

Citation

@article{madane2023transformer,
  title={Transformer-based Conditional Generative Adversarial Network for Multivariate Time Series Generation},
  author={Madane, Abdellah and Dilmi, Mohamed-djallel and Forest, Florent and Azzag, Hanane and Lebbah, Mustapha and Lacaille, Jerome},
  booktitle={International Workshop on Temporal Analytics},
  organization={Pacific-Asia Conference on Knowledge Discovery and Data Mining},
  year={2023}
}